An Intel crisis is a crisis for open virtual RANAn Intel crisis is a crisis for open virtual RAN

The hefty job cuts tearing through Intel pose a threat to the market for virtual radio access network products, which the chipmaker still dominates.

Iain Morris, International Editor

August 27, 2024

8 Min Read
Intel headquarters in the rain
(Source: Intel)

For a long time, the chips embedded within mobile network equipment, supplied by the likes of Ericsson, Huawei and Nokia, have been purpose-built for that mobile network equipment and unsuitable for much else. The dream of figures in the tech industry, and parts of the telecom sector, is to sweep these out and replace them with the sort of general-purpose central processing units (CPUs) found in data centers. Common silicon platforms and software tools would bring economies of scale and attendant cost savings. Off-site servers hosting 5G software could be used for other applications.

But this virtualization or cloudification still accounts for just 3% of the worldwide market for radio access network (RAN) products, worth approximately $40 billion last year, according to figures shared by Omdia, a Light Reading sister company. On the chips side, it is also dominated by a single player – Intel. In early 2023, the silicon behemoth boasted a 99% share of all virtual RAN deployments, and there has been little sign of change since then. This all makes the huge cuts recently announced by Intel a massive threat to this nascent virtual RAN market.

After a sequence of disappointing results, the US chipmaker early this month said it would slash about 15% of jobs, a figure that equates to more than 18,000 roles, based on headcount at the end of 2023. Its goal is to reduce operating costs to $17.5 billion next year, from about $21.7 billion in 2023, and lower capital expenditure by a fifth, compared with earlier projections, to between $25 billion and $27 billion this year. Next year, spending will fall to between $20 billion and $23 billion, Intel has said.

Yet the cuts may go even deeper at non-core units. In mid-August, CRN obtained an internal presentation that talked about reducing costs by more than 35% at the sales and marketing group, which houses its global partner organization. There is also a persistent rumor, which refuses to go away, that Intel is trying to offload some of its telecom assets.

Altera on the outside

The company has, of course, already announced the spin-off of Altera. It was the $16.7 billion acquisition of this business in 2015 that furnished Intel with much of its telecom expertise. While Intel has focused historically on the CPUs for personal computers and data-center servers, Altera specialized in a type of chip called a field programmable gate array (FPGA), which can be coded for a specific purpose after it has been shipped. Altera also made the application-specific integrated circuits (ASICs) that commonly feature in traditional basestations.

The Altera spin-off at least removes a contradictory element. In proselytizing about virtualization and cloudification, Intel must persuade skeptics that general-purpose CPUs can measure up to more customized silicon on specific tasks, such as 5G workloads. Altera is about the opposite – pushing FPGAs and ASICs as forms of accelerated computing where general purpose falls short.

Ultimately due for an initial public offering, Altera could enjoy success, as it previously did, by selling purpose-built chips for basestations. But the much bigger telecom opportunity for Intel is in converting the purpose-built industry to the virtualization faith. Altera, then, arguably has greater value as a source of telecom expertise to FlexRAN, Intel's software framework for virtual RAN. Unless there is a support contract between Altera and Intel's network and edge group, access to this expertise may be complicated by the Altera spin-off.

At the same time, and despite those rumors, Intel would struggle to sell FlexRAN to another company. This is largely because it cannot be divorced from x86, the instruction set architecture for Intel's chips. In fact, it is not even compatible with processors made by AMD, the only other big maker of x86 chips. "The way the license is written is that it can only be used on Intel processors," said Nick Hancock, the director of AMD's telco vertical, during a previous interview with Light Reading. "It is like having a car but not the keys."

Partly for this reason, FlexRAN appears to have been in retreat. Ericsson, Intel's biggest virtual RAN partner, claims to have avoided it, preferring to write virtual RAN software from scratch that it can deploy on AMD's chips without major changes (or so Ericsson says). Samsung, another big Intel customer, has similarly left FlexRAN parked on Intel's stall.

But Ericsson does use the software that Intel writes for network functions offloaded to custom silicon. For all its fanaticism about general-purpose processors, even Intel concedes that some functions need the oomph provided by accelerators, as more customized chips are called. Under the brand vRAN Boost, this Intel software has previously focused on a demanding function called forward error correction (FEC), itself a part of Layer 1, the category of RAN software that is hungriest for compute resources. Under Intel's roadmap, however, other 5G functions will need accelerating in future.

The Altera spin-off and broader cuts would seem to endanger these activities. Intel has always been relatively coy about the hardware used for its accelerators, and it did not respond to various questions asked for this article, but Ericsson formerly described them as eASICs, a hybrid of the ASICs and FPGAs that are the remit of Altera. As with Altera, Intel gained this expertise when it bought a small company using eASIC as its name in 2018 and seated its 120 employees in the Altera unit, then known as the programmable solutions group. Conversely, the software produced for the FEC and other accelerated functions is thought to reside under the same roof as FlexRAN.

Slow-moving Rapids

To critics, the growing customization of Intel's virtual RAN products seems to defeat the entire purpose. In Granite Rapids-D, a forthcoming range, accelerators and fronthaul connectivity are tightly integrated with the CPU, forming what looks to some more like a system-on-chip than general-purpose processor. The same chip would seem like overkill for more humdrum data-center needs, like employing an exterminator to flush out a single wasp.

"How is that different from any other custom chip?" railed Tommi Uitto, the head of Nokia's mobile networks business group, when he met Light Reading at this year's Mobile World Congress. "It's not a general-purpose processor. And if you build servers with only those CPUs, it would be ridiculous because you'd have all this overhead of hardware acceleration in the product cost and power consumption."

Even so, at the very heart of Granite Rapids-D is the same x86 technology found across Intel's other products. And separating any unit based on this intellectual property from the rest of Intel would be difficult if not impossible. A likelier outcome than a spin-off and sale is perhaps a shift in priorities away from virtual RAN, which has failed to make progress.

"I would say that the vRAN business has not grown during the past year for Intel," said Joe Madden of analyst firm Mobile Experts, answering questions by email. Intel's annual vRAN sales amount to just $100,000, he estimates.

Virtual RAN still looks Armless

Unfortunately, for operators persuaded of virtualization's merits, there is still no feasible commercial alternative to Intel, despite the industry's best efforts. Ericsson does not appear to have advanced beyond trials with AMD. A bigger shake-up of the market is promised by Arm, a UK-based licensor of an entirely different architecture from x86, known mainly because Arm-based chips power most of today's smartphones. But Arm licensees including Oracle-backed Ampere Computing, AWS and Nvidia have made no visible progress in virtual RAN.

Indeed, despite being a huge client of AWS, Dish Network in the US continues to rely on Intel rather than AWS's Arm-based Graviton processors for its virtual RAN. "We are still working on it," said Ishwar Parulkar, the chief telecom technologist for AWS, at the Digital Transformation World event in June. "The challenge is porting software to Arm. Today, it is primarily Intel-based, and there is work to be done, and it is engineering work."

Analysts are sanguine. While virtual RAN accounts for only 3% of the market, it represents 10% of the baseband subsector, notes Remy Pascal of Omdia, who predicts this share will double by 2028. Madden of Mobile Experts says it needs a new investment cycle. "I still believe that vRAN will eventually become the primary architecture for next-generation networks, but the world is waiting for new spectrum as a vehicle to usher in the new technology," he said via email. "Not many operators will 'rip and replace' equipment to introduce vRAN."

There is, though, a more negative possibility. Amid disappointing returns on 5G investment, many operators have shown they are in no mood for another splurge on next-generation mobile network kit. The rate of data traffic growth, a driver of network investment in the absence of profitable new applications, is slowing, say numerous analyst firms. One of them, Analysys Mason, predicts that capital intensity, or spending as a percentage of revenues, will drop sharply in the next few years.

In this environment of constrained spending, the benefits of RAN virtualization look increasingly dubious. Costs are unlikely to fall dramatically if the customization in which Intel is engaged erodes economies of scale. Compute resources could not be shared between numerous telco and IT workloads unless they were centralized to some degree in data facilities. But RAN compute happens largely at mast sites. Moving it to data centers would carry huge fronthaul costs for most operators and could have an impact on service quality.

Intel's effective stranglehold on the tiny market – worth just $1.3 billion in all revenues annually, according to Omdia's figures – is as much of an issue for chief technology officers already bothered about the lack of alternatives to Ericsson and Nokia for RAN equipment. A more competitive set-up might have persuaded a bigger number to invest by now. Unfortunately, Intel's current poor health and its control of the market have produced a toxic combination.

Read more about:

Omdia

About the Author

Iain Morris

International Editor, Light Reading

Iain Morris joined Light Reading as News Editor at the start of 2015 -- and we mean, right at the start. His friends and family were still singing Auld Lang Syne as Iain started sourcing New Year's Eve UK mobile network congestion statistics. Prior to boosting Light Reading's UK-based editorial team numbers (he is based in London, south of the river), Iain was a successful freelance writer and editor who had been covering the telecoms sector for the past 15 years. His work has appeared in publications including The Economist (classy!) and The Observer, besides a variety of trade and business journals. He was previously the lead telecoms analyst for the Economist Intelligence Unit, and before that worked as a features editor at Telecommunications magazine. Iain started out in telecoms as an editor at consulting and market-research company Analysys (now Analysys Mason).

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like